Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Use CUDA 11.2+ features via dlopen #990

Merged
merged 30 commits into from
Mar 18, 2022
Merged

Use CUDA 11.2+ features via dlopen #990

merged 30 commits into from
Mar 18, 2022

Commits on Mar 9, 2022

  1. CUDA 11.2+ features are used via dlopen 

    By binding to the cudart 11.2 functions at runtime we remove
the requirement that these symbols exist, therefore allowing
RMM to be compiled with 11.2+ and used with 11.0 or 11.1.
    @robertmaynard
    robertmaynard committed Mar 9, 2022
    Configuration menu
    Copy the full SHA
    8944f92 View commit details
    Browse the repository at this point in the history

  2. Remove cuda_async_memory_resource::is_supported

    @robertmaynard
    robertmaynard committed Mar 9, 2022
    Configuration menu
    Copy the full SHA
    838d115 View commit details
    Browse the repository at this point in the history

  3. Give the typedef used by function a better name

    @robertmaynard
    robertmaynard committed Mar 9, 2022
    Configuration menu
    Copy the full SHA
    452b998 View commit details
    Browse the repository at this point in the history

  4. Consistently use 'auto *'

    @robertmaynard
    robertmaynard committed Mar 9, 2022
    Configuration menu
    Copy the full SHA
    f05cf86 View commit details
    Browse the repository at this point in the history

Commits on Mar 10, 2022

  1. Refactor 'open_cuda_runtime' to be thread safe

    @robertmaynard
    robertmaynard committed Mar 10, 2022
    Configuration menu
    Copy the full SHA
    b070025 View commit details
    Browse the repository at this point in the history

  2. Remove an unneeded check from dynamic_load_runtime

    @robertmaynard
    robertmaynard committed Mar 10, 2022
    Configuration menu
    Copy the full SHA
    b598499 View commit details
    Browse the repository at this point in the history

  3. Correct style issues found by CI

    @robertmaynard
    robertmaynard committed Mar 10, 2022
    Configuration menu
    Copy the full SHA
    b3ff2d8 View commit details
    Browse the repository at this point in the history

  4. Refactor 'open_cuda_runtime' to 'get_cuda_runtime_handle'

    @robertmaynard
    robertmaynard committed Mar 10, 2022
    Configuration menu
    Copy the full SHA
    2b3a2b1 View commit details
    Browse the repository at this point in the history

Commits on Mar 14, 2022

  1. Ensure we call the correct cudart functions

    @robertmaynard
    robertmaynard committed Mar 14, 2022
    Configuration menu
    Copy the full SHA
    da02b55 View commit details
    Browse the repository at this point in the history

  2. Correct build failures

    @robertmaynard
    robertmaynard committed Mar 14, 2022
    Configuration menu
    Copy the full SHA
    9d228b9 View commit details
    Browse the repository at this point in the history

  3. verify RMM_SYNC_ALLOC_WRAPPER signature in static code path

    @robertmaynard
    robertmaynard committed Mar 14, 2022
    Configuration menu
    Copy the full SHA
    8780a93 View commit details
    Browse the repository at this point in the history

Commits on Mar 15, 2022

  1. Update include/rmm/detail/dynamic_load_runtime.hpp 

    Co-authored-by: Mark Harris <mharris@nvidia.com>
    @robertmaynard @harrism
    robertmaynard and harrism authored Mar 15, 2022
    Configuration menu
    Copy the full SHA
    9bfb83c View commit details
    Browse the repository at this point in the history

  2. Update include/rmm/detail/dynamic_load_runtime.hpp 

    Co-authored-by: Mark Harris <mharris@nvidia.com>
    @robertmaynard @harrism
    robertmaynard and harrism authored Mar 15, 2022
    Configuration menu
    Copy the full SHA
    2424da3 View commit details
    Browse the repository at this point in the history

  3. Update include/rmm/detail/dynamic_load_runtime.hpp 

    Co-authored-by: Mark Harris <mharris@nvidia.com>
    @robertmaynard @harrism
    robertmaynard and harrism authored Mar 15, 2022
    Configuration menu
    Copy the full SHA
    0a32c5a View commit details
    Browse the repository at this point in the history

  4. Update include/rmm/detail/dynamic_load_runtime.hpp 

    Co-authored-by: Mark Harris <mharris@nvidia.com>
    @robertmaynard @harrism
    robertmaynard and harrism authored Mar 15, 2022
    Configuration menu
    Copy the full SHA
    49df39e View commit details
    Browse the repository at this point in the history

  5. Update include/rmm/detail/dynamic_load_runtime.hpp 

    Co-authored-by: Mark Harris <mharris@nvidia.com>
    @robertmaynard @harrism
    robertmaynard and harrism authored Mar 15, 2022
    Configuration menu
    Copy the full SHA
    80e5316 View commit details
    Browse the repository at this point in the history

  6. Rename RMM_SYNC_ALLOC_WRAPPER to RMM_CUDART_API_WRAPPER

    @robertmaynard
    robertmaynard committed Mar 15, 2022
    Configuration menu
    Copy the full SHA
    8463e0f View commit details
    Browse the repository at this point in the history

  7. fix function_sig typo

    @robertmaynard
    robertmaynard committed Mar 15, 2022
    Configuration menu
    Copy the full SHA
    ff9ea91 View commit details
    Browse the repository at this point in the history

  8. Ensure we call cudaMemPoolDestroy via async_alloc

    @robertmaynard
    robertmaynard committed Mar 15, 2022
    Configuration menu
    Copy the full SHA
    a3a2cf5 View commit details
    Browse the repository at this point in the history

  9. dynamic_load_runtime now returns std::optional<function_ptr>

    @robertmaynard
    robertmaynard committed Mar 15, 2022
    Configuration menu
    Copy the full SHA
    baf618b View commit details
    Browse the repository at this point in the history

  10. dynamic_load_runtime now returns std::optional<function_ptr>

    @robertmaynard
    robertmaynard committed Mar 15, 2022
    Configuration menu
    Copy the full SHA
    132305e View commit details
    Browse the repository at this point in the history

  11. Correct style issues found by CI

    @robertmaynard
    robertmaynard committed Mar 15, 2022
    Configuration menu
    Copy the full SHA
    2ff56b7 View commit details
    Browse the repository at this point in the history

  12. return empty optional instead of optional(nullptr) 

    Co-authored-by: Jake Hemstad <jhemstad@nvidia.com>
    @robertmaynard @jrhemstad
    robertmaynard and jrhemstad authored Mar 15, 2022
    Configuration menu
    Copy the full SHA
    207cfae View commit details
    Browse the repository at this point in the history

Commits on Mar 16, 2022

  1. Correct std::optional issues found by review

    @robertmaynard
    robertmaynard committed Mar 16, 2022
    Configuration menu
    Copy the full SHA
    13ef5ea View commit details
    Browse the repository at this point in the history

  2. Correct style issues found by ci

    @robertmaynard
    robertmaynard committed Mar 16, 2022
    Configuration menu
    Copy the full SHA
    2ddc6fd View commit details
    Browse the repository at this point in the history

  3. Merge branch 'branch-22.04' into cuda_async_memory_resource-dlopen-cu… 

    …dart
    @robertmaynard
    robertmaynard committed Mar 16, 2022
    Configuration menu
    Copy the full SHA
    76b8db5 View commit details
    Browse the repository at this point in the history

Commits on Mar 17, 2022

  1. Merge branch 'branch-22.04' into cuda_async_memory_resource-dlopen-cu… 

    …dart
    @robertmaynard
    robertmaynard committed Mar 17, 2022
    Configuration menu
    Copy the full SHA
    1aad91c View commit details
    Browse the repository at this point in the history

  2. opportunistic reuse bug now updated to use rmm::detail::async_alloc

    @robertmaynard
    robertmaynard committed Mar 17, 2022
    Configuration menu
    Copy the full SHA
    1994bc4 View commit details
    Browse the repository at this point in the history

  3. Use static_assert to validate function arguments are valid

    @robertmaynard
    robertmaynard committed Mar 17, 2022
    Configuration menu
    Copy the full SHA
    48e10d6 View commit details
    Browse the repository at this point in the history

  4. remove indirect function call when building statically

    @robertmaynard
    robertmaynard committed Mar 17, 2022
    Configuration menu
    Copy the full SHA
    86ef0f9 View commit details
    Browse the repository at this point in the history